支持向量机(SVM)分类的关键是发现分类最优超平面及类别间隔,而混合像元比纯净像元更接近类别边界,更容易找出最优超平面。本文针对SVM分类器的特点,在高光谱数据分类中采用混合像元作为训练样本对SVM进行训练,试验表明采用类别边界上的混合像元作为训练样本是可行的,能够获得与纯净训练样本接近的分类精度,进一步验证了SVM分类对训练样本空间分布依赖度较低的特点。 The key to SVM classification is to find the best classification hyperplane and classification interval, while the mixed pixel is closer to the classification boundary than the pure pixel, making it easier to find the optimal hyperplane. In this paper, based on the characteristics of SVM classifier, hybrid pixels are used as training samples in hyperspectral data classification to train SVM. Experiments show that it is feasible to use mixed pixels on the boundary of classes as training samples, which can get close to pure training samples The classification accuracy of SVM classification further verifies that the SVM classification has a low dependence on the spatial distribution of training samples.